net/xfrm/xfrm_state.c

   1 /*
   2  * xfrm_state.c
   3  *
   4  * Changes:
   5  *      Mitsuru KANDA @USAGI
   6  *      Kazunori MIYAZAWA @USAGI
   7  *      Kunihiro Ishiguro <kunihiro@ipinfusion.com>
   8  *              IPv6 support
   9  *      YOSHIFUJI Hideaki @USAGI
  10  *              Split up af-specific functions
  11  *      Derek Atkins <derek@ihtfp.com>
  12  *              Add UDP Encapsulation
  13  *
  14  */
  15
  16 #include <linux/workqueue.h>
  17 #include <net/xfrm.h>
  18 #include <linux/pfkeyv2.h>
  19 #include <linux/ipsec.h>
  20 #include <linux/module.h>
  21 #include <linux/cache.h>
  22 #include <linux/audit.h>
  23 #include <linux/uaccess.h>
  24 #include <linux/ktime.h>
  25 #include <linux/slab.h>
  26 #include <linux/interrupt.h>
  27 #include <linux/kernel.h>
  28
  29 #include "xfrm_hash.h"
  30
  31 #define xfrm_state_deref_prot(table, net) \
  32         rcu_dereference_protected((table), lockdep_is_held(&(net)->xfrm.xfrm_state_lock))
  33
  34 static void xfrm_state_gc_task(struct work_struct *work);
  35
  36 /* Each xfrm_state may be linked to two tables:
  37
  38    1. Hash table by (spi,daddr,ah/esp) to find SA by SPI. (input,ctl)
  39    2. Hash table by (daddr,family,reqid) to find what SAs exist for given
  40       destination/tunnel endpoint. (output)
  41  */
  42
  43 static unsigned int xfrm_state_hashmax __read_mostly = 1 * 1024 * 1024;
  44 static __read_mostly seqcount_t xfrm_state_hash_generation = SEQCNT_ZERO(xfrm_state_hash_generation);
  45
  46 static DECLARE_WORK(xfrm_state_gc_work, xfrm_state_gc_task);
  47 static HLIST_HEAD(xfrm_state_gc_list);
  48
  49 static inline bool xfrm_state_hold_rcu(struct xfrm_state __rcu *x)
  50 {
  51         return refcount_inc_not_zero(&x->refcnt);
  52 }
  53
  54 static inline unsigned int xfrm_dst_hash(struct net *net,
  55                                          const xfrm_address_t *daddr,
  56                                          const xfrm_address_t *saddr,
  57                                          u32 reqid,
  58                                          unsigned short family)
  59 {
  60         return __xfrm_dst_hash(daddr, saddr, reqid, family, net->xfrm.state_hmask);
  61 }
  62
  63 static inline unsigned int xfrm_src_hash(struct net *net,
  64                                          const xfrm_address_t *daddr,
  65                                          const xfrm_address_t *saddr,
  66                                          unsigned short family)
  67 {
  68         return __xfrm_src_hash(daddr, saddr, family, net->xfrm.state_hmask);
  69 }
  70
  71 static inline unsigned int
  72 xfrm_spi_hash(struct net *net, const xfrm_address_t *daddr,
  73               __be32 spi, u8 proto, unsigned short family)
  74 {
  75         return __xfrm_spi_hash(daddr, spi, proto, family, net->xfrm.state_hmask);
  76 }
  77
  78 static void xfrm_hash_transfer(struct hlist_head *list,
  79                                struct hlist_head *ndsttable,
  80                                struct hlist_head *nsrctable,
  81                                struct hlist_head *nspitable,
  82                                unsigned int nhashmask)
  83 {
  84         struct hlist_node *tmp;
  85         struct xfrm_state *x;
  86
  87         hlist_for_each_entry_safe(x, tmp, list, bydst) {
  88                 unsigned int h;
  89
  90                 h = __xfrm_dst_hash(&x->id.daddr, &x->props.saddr,
  91                                     x->props.reqid, x->props.family,
  92                                     nhashmask);
  93                 hlist_add_head_rcu(&x->bydst, ndsttable + h);
  94
  95                 h = __xfrm_src_hash(&x->id.daddr, &x->props.saddr,
  96                                     x->props.family,
  97                                     nhashmask);
  98                 hlist_add_head_rcu(&x->bysrc, nsrctable + h);
  99
 100                 if (x->id.spi) {
 101                         h = __xfrm_spi_hash(&x->id.daddr, x->id.spi,
 102                                             x->id.proto, x->props.family,
 103                                             nhashmask);
 104                         hlist_add_head_rcu(&x->byspi, nspitable + h);
 105                 }
 106         }
 107 }
 108
 109 static unsigned long xfrm_hash_new_size(unsigned int state_hmask)
 110 {
 111         return ((state_hmask + 1) << 1) * sizeof(struct hlist_head);
 112 }
 113
 114 static void xfrm_hash_resize(struct work_struct *work)
 115 {
 116         struct net *net = container_of(work, struct net, xfrm.state_hash_work);
 117         struct hlist_head *ndst, *nsrc, *nspi, *odst, *osrc, *ospi;
 118         unsigned long nsize, osize;
 119         unsigned int nhashmask, ohashmask;
 120         int i;
 121
 122         nsize = xfrm_hash_new_size(net->xfrm.state_hmask);
 123         ndst = xfrm_hash_alloc(nsize);
 124         if (!ndst)
 125                 return;
 126         nsrc = xfrm_hash_alloc(nsize);
 127         if (!nsrc) {
 128                 xfrm_hash_free(ndst, nsize);
 129                 return;
 130         }
 131         nspi = xfrm_hash_alloc(nsize);
 132         if (!nspi) {
 133                 xfrm_hash_free(ndst, nsize);
 134                 xfrm_hash_free(nsrc, nsize);
 135                 return;
 136         }
 137
 138         spin_lock_bh(&net->xfrm.xfrm_state_lock);
 139         write_seqcount_begin(&xfrm_state_hash_generation);
 140
 141         nhashmask = (nsize / sizeof(struct hlist_head)) - 1U;
 142         odst = xfrm_state_deref_prot(net->xfrm.state_bydst, net);
 143         for (i = net->xfrm.state_hmask; i >= 0; i--)
 144                 xfrm_hash_transfer(odst + i, ndst, nsrc, nspi, nhashmask);
 145
 146         osrc = xfrm_state_deref_prot(net->xfrm.state_bysrc, net);
 147         ospi = xfrm_state_deref_prot(net->xfrm.state_byspi, net);
 148         ohashmask = net->xfrm.state_hmask;
 149
 150         rcu_assign_pointer(net->xfrm.state_bydst, ndst);
 151         rcu_assign_pointer(net->xfrm.state_bysrc, nsrc);
 152         rcu_assign_pointer(net->xfrm.state_byspi, nspi);
 153         net->xfrm.state_hmask = nhashmask;
 154
 155         write_seqcount_end(&xfrm_state_hash_generation);
 156         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 157
 158         osize = (ohashmask + 1) * sizeof(struct hlist_head);
 159
 160         synchronize_rcu();
 161
 162         xfrm_hash_free(odst, osize);
 163         xfrm_hash_free(osrc, osize);
 164         xfrm_hash_free(ospi, osize);
 165 }
 166
 167 static DEFINE_SPINLOCK(xfrm_state_afinfo_lock);
 168 static struct xfrm_state_afinfo __rcu *xfrm_state_afinfo[NPROTO];
 169
 170 static DEFINE_SPINLOCK(xfrm_state_gc_lock);
 171
 172 int __xfrm_state_delete(struct xfrm_state *x);
 173
 174 int km_query(struct xfrm_state *x, struct xfrm_tmpl *t, struct xfrm_policy *pol);
 175 bool km_is_alive(const struct km_event *c);
 176 void km_state_expired(struct xfrm_state *x, int hard, u32 portid);
 177
 178 static DEFINE_SPINLOCK(xfrm_type_lock);
 179 int xfrm_register_type(const struct xfrm_type *type, unsigned short family)
 180 {
 181         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
 182         const struct xfrm_type **typemap;
 183         int err = 0;
 184
 185         if (unlikely(afinfo == NULL))
 186                 return -EAFNOSUPPORT;
 187         typemap = afinfo->type_map;
 188         spin_lock_bh(&xfrm_type_lock);
 189
 190         if (likely(typemap[type->proto] == NULL))
 191                 typemap[type->proto] = type;
 192         else
 193                 err = -EEXIST;
 194         spin_unlock_bh(&xfrm_type_lock);
 195         rcu_read_unlock();
 196         return err;
 197 }
 198 EXPORT_SYMBOL(xfrm_register_type);
 199
 200 int xfrm_unregister_type(const struct xfrm_type *type, unsigned short family)
 201 {
 202         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
 203         const struct xfrm_type **typemap;
 204         int err = 0;
 205
 206         if (unlikely(afinfo == NULL))
 207                 return -EAFNOSUPPORT;
 208         typemap = afinfo->type_map;
 209         spin_lock_bh(&xfrm_type_lock);
 210
 211         if (unlikely(typemap[type->proto] != type))
 212                 err = -ENOENT;
 213         else
 214                 typemap[type->proto] = NULL;
 215         spin_unlock_bh(&xfrm_type_lock);
 216         rcu_read_unlock();
 217         return err;
 218 }
 219 EXPORT_SYMBOL(xfrm_unregister_type);
 220
 221 static const struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
 222 {
 223         struct xfrm_state_afinfo *afinfo;
 224         const struct xfrm_type **typemap;
 225         const struct xfrm_type *type;
 226         int modload_attempted = 0;
 227
 228 retry:
 229         afinfo = xfrm_state_get_afinfo(family);
 230         if (unlikely(afinfo == NULL))
 231                 return NULL;
 232         typemap = afinfo->type_map;
 233
 234         type = READ_ONCE(typemap[proto]);
 235         if (unlikely(type && !try_module_get(type->owner)))
 236                 type = NULL;
 237
 238         rcu_read_unlock();
 239
 240         if (!type && !modload_attempted) {
 241                 request_module("xfrm-type-%d-%d", family, proto);
 242                 modload_attempted = 1;
 243                 goto retry;
 244         }
 245
 246         return type;
 247 }
 248
 249 static void xfrm_put_type(const struct xfrm_type *type)
 250 {
 251         module_put(type->owner);
 252 }
 253
 254 static DEFINE_SPINLOCK(xfrm_type_offload_lock);
 255 int xfrm_register_type_offload(const struct xfrm_type_offload *type,
 256                                unsigned short family)
 257 {
 258         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
 259         const struct xfrm_type_offload **typemap;
 260         int err = 0;
 261
 262         if (unlikely(afinfo == NULL))
 263                 return -EAFNOSUPPORT;
 264         typemap = afinfo->type_offload_map;
 265         spin_lock_bh(&xfrm_type_offload_lock);
 266
 267         if (likely(typemap[type->proto] == NULL))
 268                 typemap[type->proto] = type;
 269         else
 270                 err = -EEXIST;
 271         spin_unlock_bh(&xfrm_type_offload_lock);
 272         rcu_read_unlock();
 273         return err;
 274 }
 275 EXPORT_SYMBOL(xfrm_register_type_offload);
 276
 277 int xfrm_unregister_type_offload(const struct xfrm_type_offload *type,
 278                                  unsigned short family)
 279 {
 280         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
 281         const struct xfrm_type_offload **typemap;
 282         int err = 0;
 283
 284         if (unlikely(afinfo == NULL))
 285                 return -EAFNOSUPPORT;
 286         typemap = afinfo->type_offload_map;
 287         spin_lock_bh(&xfrm_type_offload_lock);
 288
 289         if (unlikely(typemap[type->proto] != type))
 290                 err = -ENOENT;
 291         else
 292                 typemap[type->proto] = NULL;
 293         spin_unlock_bh(&xfrm_type_offload_lock);
 294         rcu_read_unlock();
 295         return err;
 296 }
 297 EXPORT_SYMBOL(xfrm_unregister_type_offload);
 298
 299 static const struct xfrm_type_offload *
 300 xfrm_get_type_offload(u8 proto, unsigned short family, bool try_load)
 301 {
 302         struct xfrm_state_afinfo *afinfo;
 303         const struct xfrm_type_offload **typemap;
 304         const struct xfrm_type_offload *type;
 305
 306 retry:
 307         afinfo = xfrm_state_get_afinfo(family);
 308         if (unlikely(afinfo == NULL))
 309                 return NULL;
 310         typemap = afinfo->type_offload_map;
 311
 312         type = typemap[proto];
 313         if ((type && !try_module_get(type->owner)))
 314                 type = NULL;
 315
 316         if (!type && try_load) {
 317                 request_module("xfrm-offload-%d-%d", family, proto);
 318                 try_load = 0;
 319                 goto retry;
 320         }
 321
 322         rcu_read_unlock();
 323         return type;
 324 }
 325
 326 static void xfrm_put_type_offload(const struct xfrm_type_offload *type)
 327 {
 328         module_put(type->owner);
 329 }
 330
 331 static DEFINE_SPINLOCK(xfrm_mode_lock);
 332 int xfrm_register_mode(struct xfrm_mode *mode, int family)
 333 {
 334         struct xfrm_state_afinfo *afinfo;
 335         struct xfrm_mode **modemap;
 336         int err;
 337
 338         if (unlikely(mode->encap >= XFRM_MODE_MAX))
 339                 return -EINVAL;
 340
 341         afinfo = xfrm_state_get_afinfo(family);
 342         if (unlikely(afinfo == NULL))
 343                 return -EAFNOSUPPORT;
 344
 345         err = -EEXIST;
 346         modemap = afinfo->mode_map;
 347         spin_lock_bh(&xfrm_mode_lock);
 348         if (modemap[mode->encap])
 349                 goto out;
 350
 351         err = -ENOENT;
 352         if (!try_module_get(afinfo->owner))
 353                 goto out;
 354
 355         mode->afinfo = afinfo;
 356         modemap[mode->encap] = mode;
 357         err = 0;
 358
 359 out:
 360         spin_unlock_bh(&xfrm_mode_lock);
 361         rcu_read_unlock();
 362         return err;
 363 }
 364 EXPORT_SYMBOL(xfrm_register_mode);
 365
 366 int xfrm_unregister_mode(struct xfrm_mode *mode, int family)
 367 {
 368         struct xfrm_state_afinfo *afinfo;
 369         struct xfrm_mode **modemap;
 370         int err;
 371
 372         if (unlikely(mode->encap >= XFRM_MODE_MAX))
 373                 return -EINVAL;
 374
 375         afinfo = xfrm_state_get_afinfo(family);
 376         if (unlikely(afinfo == NULL))
 377                 return -EAFNOSUPPORT;
 378
 379         err = -ENOENT;
 380         modemap = afinfo->mode_map;
 381         spin_lock_bh(&xfrm_mode_lock);
 382         if (likely(modemap[mode->encap] == mode)) {
 383                 modemap[mode->encap] = NULL;
 384                 module_put(mode->afinfo->owner);
 385                 err = 0;
 386         }
 387
 388         spin_unlock_bh(&xfrm_mode_lock);
 389         rcu_read_unlock();
 390         return err;
 391 }
 392 EXPORT_SYMBOL(xfrm_unregister_mode);
 393
 394 static struct xfrm_mode *xfrm_get_mode(unsigned int encap, int family)
 395 {
 396         struct xfrm_state_afinfo *afinfo;
 397         struct xfrm_mode *mode;
 398         int modload_attempted = 0;
 399
 400         if (unlikely(encap >= XFRM_MODE_MAX))
 401                 return NULL;
 402
 403 retry:
 404         afinfo = xfrm_state_get_afinfo(family);
 405         if (unlikely(afinfo == NULL))
 406                 return NULL;
 407
 408         mode = READ_ONCE(afinfo->mode_map[encap]);
 409         if (unlikely(mode && !try_module_get(mode->owner)))
 410                 mode = NULL;
 411
 412         rcu_read_unlock();
 413         if (!mode && !modload_attempted) {
 414                 request_module("xfrm-mode-%d-%d", family, encap);
 415                 modload_attempted = 1;
 416                 goto retry;
 417         }
 418
 419         return mode;
 420 }
 421
 422 static void xfrm_put_mode(struct xfrm_mode *mode)
 423 {
 424         module_put(mode->owner);
 425 }
 426
 427 static void xfrm_state_gc_destroy(struct xfrm_state *x)
 428 {
 429         tasklet_hrtimer_cancel(&x->mtimer);
 430         del_timer_sync(&x->rtimer);
 431         kfree(x->aead);
 432         kfree(x->aalg);
 433         kfree(x->ealg);
 434         kfree(x->calg);
 435         kfree(x->encap);
 436         kfree(x->coaddr);
 437         kfree(x->replay_esn);
 438         kfree(x->preplay_esn);
 439         if (x->inner_mode)
 440                 xfrm_put_mode(x->inner_mode);
 441         if (x->inner_mode_iaf)
 442                 xfrm_put_mode(x->inner_mode_iaf);
 443         if (x->outer_mode)
 444                 xfrm_put_mode(x->outer_mode);
 445         if (x->type_offload)
 446                 xfrm_put_type_offload(x->type_offload);
 447         if (x->type) {
 448                 x->type->destructor(x);
 449                 xfrm_put_type(x->type);
 450         }
 451         xfrm_dev_state_free(x);
 452         security_xfrm_state_free(x);
 453         kfree(x);
 454 }
 455
 456 static void xfrm_state_gc_task(struct work_struct *work)
 457 {
 458         struct xfrm_state *x;
 459         struct hlist_node *tmp;
 460         struct hlist_head gc_list;
 461
 462         spin_lock_bh(&xfrm_state_gc_lock);
 463         hlist_move_list(&xfrm_state_gc_list, &gc_list);
 464         spin_unlock_bh(&xfrm_state_gc_lock);
 465
 466         synchronize_rcu();
 467
 468         hlist_for_each_entry_safe(x, tmp, &gc_list, gclist)
 469                 xfrm_state_gc_destroy(x);
 470 }
 471
 472 static enum hrtimer_restart xfrm_timer_handler(struct hrtimer *me)
 473 {
 474         struct tasklet_hrtimer *thr = container_of(me, struct tasklet_hrtimer, timer);
 475         struct xfrm_state *x = container_of(thr, struct xfrm_state, mtimer);
 476         unsigned long now = get_seconds();
 477         long next = LONG_MAX;
 478         int warn = 0;
 479         int err = 0;
 480
 481         spin_lock(&x->lock);
 482         if (x->km.state == XFRM_STATE_DEAD)
 483                 goto out;
 484         if (x->km.state == XFRM_STATE_EXPIRED)
 485                 goto expired;
 486         if (x->lft.hard_add_expires_seconds) {
 487                 long tmo = x->lft.hard_add_expires_seconds +
 488                         x->curlft.add_time - now;
 489                 if (tmo <= 0) {
 490                         if (x->xflags & XFRM_SOFT_EXPIRE) {
 491                                 /* enter hard expire without soft expire first?!
 492                                  * setting a new date could trigger this.
 493                                  * workaround: fix x->curflt.add_time by below:
 494                                  */
 495                                 x->curlft.add_time = now - x->saved_tmo - 1;
 496                                 tmo = x->lft.hard_add_expires_seconds - x->saved_tmo;
 497                         } else
 498                                 goto expired;
 499                 }
 500                 if (tmo < next)
 501                         next = tmo;
 502         }
 503         if (x->lft.hard_use_expires_seconds) {
 504                 long tmo = x->lft.hard_use_expires_seconds +
 505                         (x->curlft.use_time ? : now) - now;
 506                 if (tmo <= 0)
 507                         goto expired;
 508                 if (tmo < next)
 509                         next = tmo;
 510         }
 511         if (x->km.dying)
 512                 goto resched;
 513         if (x->lft.soft_add_expires_seconds) {
 514                 long tmo = x->lft.soft_add_expires_seconds +
 515                         x->curlft.add_time - now;
 516                 if (tmo <= 0) {
 517                         warn = 1;
 518                         x->xflags &= ~XFRM_SOFT_EXPIRE;
 519                 } else if (tmo < next) {
 520                         next = tmo;
 521                         x->xflags |= XFRM_SOFT_EXPIRE;
 522                         x->saved_tmo = tmo;
 523                 }
 524         }
 525         if (x->lft.soft_use_expires_seconds) {
 526                 long tmo = x->lft.soft_use_expires_seconds +
 527                         (x->curlft.use_time ? : now) - now;
 528                 if (tmo <= 0)
 529                         warn = 1;
 530                 else if (tmo < next)
 531                         next = tmo;
 532         }
 533
 534         x->km.dying = warn;
 535         if (warn)
 536                 km_state_expired(x, 0, 0);
 537 resched:
 538         if (next != LONG_MAX) {
 539                 tasklet_hrtimer_start(&x->mtimer, ktime_set(next, 0), HRTIMER_MODE_REL);
 540         }
 541
 542         goto out;
 543
 544 expired:
 545         if (x->km.state == XFRM_STATE_ACQ && x->id.spi == 0)
 546                 x->km.state = XFRM_STATE_EXPIRED;
 547
 548         err = __xfrm_state_delete(x);
 549         if (!err)
 550                 km_state_expired(x, 1, 0);
 551
 552         xfrm_audit_state_delete(x, err ? 0 : 1, true);
 553
 554 out:
 555         spin_unlock(&x->lock);
 556         return HRTIMER_NORESTART;
 557 }
 558
 559 static void xfrm_replay_timer_handler(unsigned long data);
 560
 561 struct xfrm_state *xfrm_state_alloc(struct net *net)
 562 {
 563         struct xfrm_state *x;
 564
 565         x = kzalloc(sizeof(struct xfrm_state), GFP_ATOMIC);
 566
 567         if (x) {
 568                 write_pnet(&x->xs_net, net);
 569                 refcount_set(&x->refcnt, 1);
 570                 atomic_set(&x->tunnel_users, 0);
 571                 INIT_LIST_HEAD(&x->km.all);
 572                 INIT_HLIST_NODE(&x->bydst);
 573                 INIT_HLIST_NODE(&x->bysrc);
 574                 INIT_HLIST_NODE(&x->byspi);
 575                 tasklet_hrtimer_init(&x->mtimer, xfrm_timer_handler,
 576                                         CLOCK_BOOTTIME, HRTIMER_MODE_ABS);
 577                 setup_timer(&x->rtimer, xfrm_replay_timer_handler,
 578                                 (unsigned long)x);
 579                 x->curlft.add_time = get_seconds();
 580                 x->lft.soft_byte_limit = XFRM_INF;
 581                 x->lft.soft_packet_limit = XFRM_INF;
 582                 x->lft.hard_byte_limit = XFRM_INF;
 583                 x->lft.hard_packet_limit = XFRM_INF;
 584                 x->replay_maxage = 0;
 585                 x->replay_maxdiff = 0;
 586                 x->inner_mode = NULL;
 587                 x->inner_mode_iaf = NULL;
 588                 spin_lock_init(&x->lock);
 589         }
 590         return x;
 591 }
 592 EXPORT_SYMBOL(xfrm_state_alloc);
 593
 594 void __xfrm_state_destroy(struct xfrm_state *x)
 595 {
 596         WARN_ON(x->km.state != XFRM_STATE_DEAD);
 597
 598         spin_lock_bh(&xfrm_state_gc_lock);
 599         hlist_add_head(&x->gclist, &xfrm_state_gc_list);
 600         spin_unlock_bh(&xfrm_state_gc_lock);
 601         schedule_work(&xfrm_state_gc_work);
 602 }
 603 EXPORT_SYMBOL(__xfrm_state_destroy);
 604
 605 int __xfrm_state_delete(struct xfrm_state *x)
 606 {
 607         struct net *net = xs_net(x);
 608         int err = -ESRCH;
 609
 610         if (x->km.state != XFRM_STATE_DEAD) {
 611                 x->km.state = XFRM_STATE_DEAD;
 612                 spin_lock(&net->xfrm.xfrm_state_lock);
 613                 list_del(&x->km.all);
 614                 hlist_del_rcu(&x->bydst);
 615                 hlist_del_rcu(&x->bysrc);
 616                 if (x->id.spi)
 617                         hlist_del_rcu(&x->byspi);
 618                 net->xfrm.state_num--;
 619                 spin_unlock(&net->xfrm.xfrm_state_lock);
 620
 621                 xfrm_dev_state_delete(x);
 622
 623                 /* All xfrm_state objects are created by xfrm_state_alloc.
 624                  * The xfrm_state_alloc call gives a reference, and that
 625                  * is what we are dropping here.
 626                  */
 627                 xfrm_state_put(x);
 628                 err = 0;
 629         }
 630
 631         return err;
 632 }
 633 EXPORT_SYMBOL(__xfrm_state_delete);
 634
 635 int xfrm_state_delete(struct xfrm_state *x)
 636 {
 637         int err;
 638
 639         spin_lock_bh(&x->lock);
 640         err = __xfrm_state_delete(x);
 641         spin_unlock_bh(&x->lock);
 642
 643         return err;
 644 }
 645 EXPORT_SYMBOL(xfrm_state_delete);
 646
 647 #ifdef CONFIG_SECURITY_NETWORK_XFRM
 648 static inline int
 649 xfrm_state_flush_secctx_check(struct net *net, u8 proto, bool task_valid)
 650 {
 651         int i, err = 0;
 652
 653         for (i = 0; i <= net->xfrm.state_hmask; i++) {
 654                 struct xfrm_state *x;
 655
 656                 hlist_for_each_entry(x, net->xfrm.state_bydst+i, bydst) {
 657                         if (xfrm_id_proto_match(x->id.proto, proto) &&
 658                            (err = security_xfrm_state_delete(x)) != 0) {
 659                                 xfrm_audit_state_delete(x, 0, task_valid);
 660                                 return err;
 661                         }
 662                 }
 663         }
 664
 665         return err;
 666 }
 667
 668 static inline int
 669 xfrm_dev_state_flush_secctx_check(struct net *net, struct net_device *dev, bool task_valid)
 670 {
 671         int i, err = 0;
 672
 673         for (i = 0; i <= net->xfrm.state_hmask; i++) {
 674                 struct xfrm_state *x;
 675                 struct xfrm_state_offload *xso;
 676
 677                 hlist_for_each_entry(x, net->xfrm.state_bydst+i, bydst) {
 678                         xso = &x->xso;
 679
 680                         if (xso->dev == dev &&
 681                            (err = security_xfrm_state_delete(x)) != 0) {
 682                                 xfrm_audit_state_delete(x, 0, task_valid);
 683                                 return err;
 684                         }
 685                 }
 686         }
 687
 688         return err;
 689 }
 690 #else
 691 static inline int
 692 xfrm_state_flush_secctx_check(struct net *net, u8 proto, bool task_valid)
 693 {
 694         return 0;
 695 }
 696
 697 static inline int
 698 xfrm_dev_state_flush_secctx_check(struct net *net, struct net_device *dev, bool task_valid)
 699 {
 700         return 0;
 701 }
 702 #endif
 703
 704 int xfrm_state_flush(struct net *net, u8 proto, bool task_valid)
 705 {
 706         int i, err = 0, cnt = 0;
 707
 708         spin_lock_bh(&net->xfrm.xfrm_state_lock);
 709         err = xfrm_state_flush_secctx_check(net, proto, task_valid);
 710         if (err)
 711                 goto out;
 712
 713         err = -ESRCH;
 714         for (i = 0; i <= net->xfrm.state_hmask; i++) {
 715                 struct xfrm_state *x;
 716 restart:
 717                 hlist_for_each_entry(x, net->xfrm.state_bydst+i, bydst) {
 718                         if (!xfrm_state_kern(x) &&
 719                             xfrm_id_proto_match(x->id.proto, proto)) {
 720                                 xfrm_state_hold(x);
 721                                 spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 722
 723                                 err = xfrm_state_delete(x);
 724                                 xfrm_audit_state_delete(x, err ? 0 : 1,
 725                                                         task_valid);
 726                                 xfrm_state_put(x);
 727                                 if (!err)
 728                                         cnt++;
 729
 730                                 spin_lock_bh(&net->xfrm.xfrm_state_lock);
 731                                 goto restart;
 732                         }
 733                 }
 734         }
 735         if (cnt) {
 736                 err = 0;
 737                 xfrm_policy_cache_flush();
 738         }
 739 out:
 740         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 741         return err;
 742 }
 743 EXPORT_SYMBOL(xfrm_state_flush);
 744
 745 int xfrm_dev_state_flush(struct net *net, struct net_device *dev, bool task_valid)
 746 {
 747         int i, err = 0, cnt = 0;
 748
 749         spin_lock_bh(&net->xfrm.xfrm_state_lock);
 750         err = xfrm_dev_state_flush_secctx_check(net, dev, task_valid);
 751         if (err)
 752                 goto out;
 753
 754         err = -ESRCH;
 755         for (i = 0; i <= net->xfrm.state_hmask; i++) {
 756                 struct xfrm_state *x;
 757                 struct xfrm_state_offload *xso;
 758 restart:
 759                 hlist_for_each_entry(x, net->xfrm.state_bydst+i, bydst) {
 760                         xso = &x->xso;
 761
 762                         if (!xfrm_state_kern(x) && xso->dev == dev) {
 763                                 xfrm_state_hold(x);
 764                                 spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 765
 766                                 err = xfrm_state_delete(x);
 767                                 xfrm_audit_state_delete(x, err ? 0 : 1,
 768                                                         task_valid);
 769                                 xfrm_state_put(x);
 770                                 if (!err)
 771                                         cnt++;
 772
 773                                 spin_lock_bh(&net->xfrm.xfrm_state_lock);
 774                                 goto restart;
 775                         }
 776                 }
 777         }
 778         if (cnt)
 779                 err = 0;
 780
 781 out:
 782         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 783         return err;
 784 }
 785 EXPORT_SYMBOL(xfrm_dev_state_flush);
 786
 787 void xfrm_sad_getinfo(struct net *net, struct xfrmk_sadinfo *si)
 788 {
 789         spin_lock_bh(&net->xfrm.xfrm_state_lock);
 790         si->sadcnt = net->xfrm.state_num;
 791         si->sadhcnt = net->xfrm.state_hmask;
 792         si->sadhmcnt = xfrm_state_hashmax;
 793         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 794 }
 795 EXPORT_SYMBOL(xfrm_sad_getinfo);
 796
 797 static void
 798 xfrm_init_tempstate(struct xfrm_state *x, const struct flowi *fl,
 799                     const struct xfrm_tmpl *tmpl,
 800                     const xfrm_address_t *daddr, const xfrm_address_t *saddr,
 801                     unsigned short family)
 802 {
 803         struct xfrm_state_afinfo *afinfo = xfrm_state_afinfo_get_rcu(family);
 804
 805         if (!afinfo)
 806                 return;
 807
 808         afinfo->init_tempsel(&x->sel, fl);
 809
 810         if (family != tmpl->encap_family) {
 811                 afinfo = xfrm_state_afinfo_get_rcu(tmpl->encap_family);
 812                 if (!afinfo)
 813                         return;
 814         }
 815         afinfo->init_temprop(x, tmpl, daddr, saddr);
 816 }
 817
 818 static struct xfrm_state *__xfrm_state_lookup(struct net *net, u32 mark,
 819                                               const xfrm_address_t *daddr,
 820                                               __be32 spi, u8 proto,
 821                                               unsigned short family)
 822 {
 823         unsigned int h = xfrm_spi_hash(net, daddr, spi, proto, family);
 824         struct xfrm_state *x;
 825
 826         hlist_for_each_entry_rcu(x, net->xfrm.state_byspi + h, byspi) {
 827                 if (x->props.family != family ||
 828                     x->id.spi       != spi ||
 829                     x->id.proto     != proto ||
 830                     !xfrm_addr_equal(&x->id.daddr, daddr, family))
 831                         continue;
 832
 833                 if ((mark & x->mark.m) != x->mark.v)
 834                         continue;
 835                 if (!xfrm_state_hold_rcu(x))
 836                         continue;
 837                 return x;
 838         }
 839
 840         return NULL;
 841 }
 842
 843 static struct xfrm_state *__xfrm_state_lookup_byaddr(struct net *net, u32 mark,
 844                                                      const xfrm_address_t *daddr,
 845                                                      const xfrm_address_t *saddr,
 846                                                      u8 proto, unsigned short family)
 847 {
 848         unsigned int h = xfrm_src_hash(net, daddr, saddr, family);
 849         struct xfrm_state *x;
 850
 851         hlist_for_each_entry_rcu(x, net->xfrm.state_bysrc + h, bysrc) {
 852                 if (x->props.family != family ||
 853                     x->id.proto     != proto ||
 854                     !xfrm_addr_equal(&x->id.daddr, daddr, family) ||
 855                     !xfrm_addr_equal(&x->props.saddr, saddr, family))
 856                         continue;
 857
 858                 if ((mark & x->mark.m) != x->mark.v)
 859                         continue;
 860                 if (!xfrm_state_hold_rcu(x))
 861                         continue;
 862                 return x;
 863         }
 864
 865         return NULL;
 866 }
 867
 868 static inline struct xfrm_state *
 869 __xfrm_state_locate(struct xfrm_state *x, int use_spi, int family)
 870 {
 871         struct net *net = xs_net(x);
 872         u32 mark = x->mark.v & x->mark.m;
 873
 874         if (use_spi)
 875                 return __xfrm_state_lookup(net, mark, &x->id.daddr,
 876                                            x->id.spi, x->id.proto, family);
 877         else
 878                 return __xfrm_state_lookup_byaddr(net, mark,
 879                                                   &x->id.daddr,
 880                                                   &x->props.saddr,
 881                                                   x->id.proto, family);
 882 }
 883
 884 static void xfrm_hash_grow_check(struct net *net, int have_hash_collision)
 885 {
 886         if (have_hash_collision &&
 887             (net->xfrm.state_hmask + 1) < xfrm_state_hashmax &&
 888             net->xfrm.state_num > net->xfrm.state_hmask)
 889                 schedule_work(&net->xfrm.state_hash_work);
 890 }
 891
 892 static void xfrm_state_look_at(struct xfrm_policy *pol, struct xfrm_state *x,
 893                                const struct flowi *fl, unsigned short family,
 894                                struct xfrm_state **best, int *acq_in_progress,
 895                                int *error)
 896 {
 897         /* Resolution logic:
 898          * 1. There is a valid state with matching selector. Done.
 899          * 2. Valid state with inappropriate selector. Skip.
 900          *
 901          * Entering area of "sysdeps".
 902          *
 903          * 3. If state is not valid, selector is temporary, it selects
 904          *    only session which triggered previous resolution. Key
 905          *    manager will do something to install a state with proper
 906          *    selector.
 907          */
 908         if (x->km.state == XFRM_STATE_VALID) {
 909                 if ((x->sel.family &&
 910                      !xfrm_selector_match(&x->sel, fl, x->sel.family)) ||
 911                     !security_xfrm_state_pol_flow_match(x, pol, fl))
 912                         return;
 913
 914                 if (!*best ||
 915                     (*best)->km.dying > x->km.dying ||
 916                     ((*best)->km.dying == x->km.dying &&
 917                      (*best)->curlft.add_time < x->curlft.add_time))
 918                         *best = x;
 919         } else if (x->km.state == XFRM_STATE_ACQ) {
 920                 *acq_in_progress = 1;
 921         } else if (x->km.state == XFRM_STATE_ERROR ||
 922                    x->km.state == XFRM_STATE_EXPIRED) {
 923                 if (xfrm_selector_match(&x->sel, fl, x->sel.family) &&
 924                     security_xfrm_state_pol_flow_match(x, pol, fl))
 925                         *error = -ESRCH;
 926         }
 927 }
 928
 929 struct xfrm_state *
 930 xfrm_state_find(const xfrm_address_t *daddr, const xfrm_address_t *saddr,
 931                 const struct flowi *fl, struct xfrm_tmpl *tmpl,
 932                 struct xfrm_policy *pol, int *err,
 933                 unsigned short family)
 934 {
 935         static xfrm_address_t saddr_wildcard = { };
 936         struct net *net = xp_net(pol);
 937         unsigned int h, h_wildcard;
 938         struct xfrm_state *x, *x0, *to_put;
 939         int acquire_in_progress = 0;
 940         int error = 0;
 941         struct xfrm_state *best = NULL;
 942         u32 mark = pol->mark.v & pol->mark.m;
 943         unsigned short encap_family = tmpl->encap_family;
 944         unsigned int sequence;
 945         struct km_event c;
 946
 947         to_put = NULL;
 948
 949         sequence = read_seqcount_begin(&xfrm_state_hash_generation);
 950
 951         rcu_read_lock();
 952         h = xfrm_dst_hash(net, daddr, saddr, tmpl->reqid, encap_family);
 953         hlist_for_each_entry_rcu(x, net->xfrm.state_bydst + h, bydst) {
 954                 if (x->props.family == encap_family &&
 955                     x->props.reqid == tmpl->reqid &&
 956                     (mark & x->mark.m) == x->mark.v &&
 957                     !(x->props.flags & XFRM_STATE_WILDRECV) &&
 958                     xfrm_state_addr_check(x, daddr, saddr, encap_family) &&
 959                     tmpl->mode == x->props.mode &&
 960                     tmpl->id.proto == x->id.proto &&
 961                     (tmpl->id.spi == x->id.spi || !tmpl->id.spi))
 962                         xfrm_state_look_at(pol, x, fl, encap_family,
 963                                            &best, &acquire_in_progress, &error);
 964         }
 965         if (best || acquire_in_progress)
 966                 goto found;
 967
 968         h_wildcard = xfrm_dst_hash(net, daddr, &saddr_wildcard, tmpl->reqid, encap_family);
 969         hlist_for_each_entry_rcu(x, net->xfrm.state_bydst + h_wildcard, bydst) {
 970                 if (x->props.family == encap_family &&
 971                     x->props.reqid == tmpl->reqid &&
 972                     (mark & x->mark.m) == x->mark.v &&
 973                     !(x->props.flags & XFRM_STATE_WILDRECV) &&
 974                     xfrm_addr_equal(&x->id.daddr, daddr, encap_family) &&
 975                     tmpl->mode == x->props.mode &&
 976                     tmpl->id.proto == x->id.proto &&
 977                     (tmpl->id.spi == x->id.spi || !tmpl->id.spi))
 978                         xfrm_state_look_at(pol, x, fl, encap_family,
 979                                            &best, &acquire_in_progress, &error);
 980         }
 981
 982 found:
 983         x = best;
 984         if (!x && !error && !acquire_in_progress) {
 985                 if (tmpl->id.spi &&
 986                     (x0 = __xfrm_state_lookup(net, mark, daddr, tmpl->id.spi,
 987                                               tmpl->id.proto, encap_family)) != NULL) {
 988                         to_put = x0;
 989                         error = -EEXIST;
 990                         goto out;
 991                 }
 992
 993                 c.net = net;
 994                 /* If the KMs have no listeners (yet...), avoid allocating an SA
 995                  * for each and every packet - garbage collection might not
 996                  * handle the flood.
 997                  */
 998                 if (!km_is_alive(&c)) {
 999                         error = -ESRCH;
1000                         goto out;
1001                 }
1002
1003                 x = xfrm_state_alloc(net);
1004                 if (x == NULL) {
1005                         error = -ENOMEM;
1006                         goto out;
1007                 }
1008                 /* Initialize temporary state matching only
1009                  * to current session. */
1010                 xfrm_init_tempstate(x, fl, tmpl, daddr, saddr, family);
1011                 memcpy(&x->mark, &pol->mark, sizeof(x->mark));
1012
1013                 error = security_xfrm_state_alloc_acquire(x, pol->security, fl->flowi_secid);
1014                 if (error) {
1015                         x->km.state = XFRM_STATE_DEAD;
1016                         to_put = x;
1017                         x = NULL;
1018                         goto out;
1019                 }
1020
1021                 if (km_query(x, tmpl, pol) == 0) {
1022                         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1023                         x->km.state = XFRM_STATE_ACQ;
1024                         list_add(&x->km.all, &net->xfrm.state_all);
1025                         hlist_add_head_rcu(&x->bydst, net->xfrm.state_bydst + h);
1026                         h = xfrm_src_hash(net, daddr, saddr, encap_family);
1027                         hlist_add_head_rcu(&x->bysrc, net->xfrm.state_bysrc + h);
1028                         if (x->id.spi) {
1029                                 h = xfrm_spi_hash(net, &x->id.daddr, x->id.spi, x->id.proto, encap_family);
1030                                 hlist_add_head_rcu(&x->byspi, net->xfrm.state_byspi + h);
1031                         }
1032                         x->lft.hard_add_expires_seconds = net->xfrm.sysctl_acq_expires;
1033                         tasklet_hrtimer_start(&x->mtimer, ktime_set(net->xfrm.sysctl_acq_expires, 0), HRTIMER_MODE_REL);
1034                         net->xfrm.state_num++;
1035                         xfrm_hash_grow_check(net, x->bydst.next != NULL);
1036                         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1037                 } else {
1038                         x->km.state = XFRM_STATE_DEAD;
1039                         to_put = x;
1040                         x = NULL;
1041                         error = -ESRCH;
1042                 }
1043         }
1044 out:
1045         if (x) {
1046                 if (!xfrm_state_hold_rcu(x)) {
1047                         *err = -EAGAIN;
1048                         x = NULL;
1049                 }
1050         } else {
1051                 *err = acquire_in_progress ? -EAGAIN : error;
1052         }
1053         rcu_read_unlock();
1054         if (to_put)
1055                 xfrm_state_put(to_put);
1056
1057         if (read_seqcount_retry(&xfrm_state_hash_generation, sequence)) {
1058                 *err = -EAGAIN;
1059                 if (x) {
1060                         xfrm_state_put(x);
1061                         x = NULL;
1062                 }
1063         }
1064
1065         return x;
1066 }
1067
1068 struct xfrm_state *
1069 xfrm_stateonly_find(struct net *net, u32 mark,
1070                     xfrm_address_t *daddr, xfrm_address_t *saddr,
1071                     unsigned short family, u8 mode, u8 proto, u32 reqid)
1072 {
1073         unsigned int h;
1074         struct xfrm_state *rx = NULL, *x = NULL;
1075
1076         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1077         h = xfrm_dst_hash(net, daddr, saddr, reqid, family);
1078         hlist_for_each_entry(x, net->xfrm.state_bydst+h, bydst) {
1079                 if (x->props.family == family &&
1080                     x->props.reqid == reqid &&
1081                     (mark & x->mark.m) == x->mark.v &&
1082                     !(x->props.flags & XFRM_STATE_WILDRECV) &&
1083                     xfrm_state_addr_check(x, daddr, saddr, family) &&
1084                     mode == x->props.mode &&
1085                     proto == x->id.proto &&
1086                     x->km.state == XFRM_STATE_VALID) {
1087                         rx = x;
1088                         break;
1089                 }
1090         }
1091
1092         if (rx)
1093                 xfrm_state_hold(rx);
1094         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1095
1096
1097         return rx;
1098 }
1099 EXPORT_SYMBOL(xfrm_stateonly_find);
1100
1101 struct xfrm_state *xfrm_state_lookup_byspi(struct net *net, __be32 spi,
1102                                               unsigned short family)
1103 {
1104         struct xfrm_state *x;
1105         struct xfrm_state_walk *w;
1106
1107         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1108         list_for_each_entry(w, &net->xfrm.state_all, all) {
1109                 x = container_of(w, struct xfrm_state, km);
1110                 if (x->props.family != family ||
1111                         x->id.spi != spi)
1112                         continue;
1113
1114                 xfrm_state_hold(x);
1115                 spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1116                 return x;
1117         }
1118         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1119         return NULL;
1120 }
1121 EXPORT_SYMBOL(xfrm_state_lookup_byspi);
1122
1123 static void __xfrm_state_insert(struct xfrm_state *x)
1124 {
1125         struct net *net = xs_net(x);
1126         unsigned int h;
1127
1128         list_add(&x->km.all, &net->xfrm.state_all);
1129
1130         h = xfrm_dst_hash(net, &x->id.daddr, &x->props.saddr,
1131                           x->props.reqid, x->props.family);
1132         hlist_add_head_rcu(&x->bydst, net->xfrm.state_bydst + h);
1133
1134         h = xfrm_src_hash(net, &x->id.daddr, &x->props.saddr, x->props.family);
1135         hlist_add_head_rcu(&x->bysrc, net->xfrm.state_bysrc + h);
1136
1137         if (x->id.spi) {
1138                 h = xfrm_spi_hash(net, &x->id.daddr, x->id.spi, x->id.proto,
1139                                   x->props.family);
1140
1141                 hlist_add_head_rcu(&x->byspi, net->xfrm.state_byspi + h);
1142         }
1143
1144         tasklet_hrtimer_start(&x->mtimer, ktime_set(1, 0), HRTIMER_MODE_REL);
1145         if (x->replay_maxage)
1146                 mod_timer(&x->rtimer, jiffies + x->replay_maxage);
1147
1148         net->xfrm.state_num++;
1149
1150         xfrm_hash_grow_check(net, x->bydst.next != NULL);
1151 }
1152
1153 /* net->xfrm.xfrm_state_lock is held */
1154 static void __xfrm_state_bump_genids(struct xfrm_state *xnew)
1155 {
1156         struct net *net = xs_net(xnew);
1157         unsigned short family = xnew->props.family;
1158         u32 reqid = xnew->props.reqid;
1159         struct xfrm_state *x;
1160         unsigned int h;
1161         u32 mark = xnew->mark.v & xnew->mark.m;
1162
1163         h = xfrm_dst_hash(net, &xnew->id.daddr, &xnew->props.saddr, reqid, family);
1164         hlist_for_each_entry(x, net->xfrm.state_bydst+h, bydst) {
1165                 if (x->props.family     == family &&
1166                     x->props.reqid      == reqid &&
1167                     (mark & x->mark.m) == x->mark.v &&
1168                     xfrm_addr_equal(&x->id.daddr, &xnew->id.daddr, family) &&
1169                     xfrm_addr_equal(&x->props.saddr, &xnew->props.saddr, family))
1170                         x->genid++;
1171         }
1172 }
1173
1174 void xfrm_state_insert(struct xfrm_state *x)
1175 {
1176         struct net *net = xs_net(x);
1177
1178         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1179         __xfrm_state_bump_genids(x);
1180         __xfrm_state_insert(x);
1181         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1182 }
1183 EXPORT_SYMBOL(xfrm_state_insert);
1184
1185 /* net->xfrm.xfrm_state_lock is held */
1186 static struct xfrm_state *__find_acq_core(struct net *net,
1187                                           const struct xfrm_mark *m,
1188                                           unsigned short family, u8 mode,
1189                                           u32 reqid, u8 proto,
1190                                           const xfrm_address_t *daddr,
1191                                           const xfrm_address_t *saddr,
1192                                           int create)
1193 {
1194         unsigned int h = xfrm_dst_hash(net, daddr, saddr, reqid, family);
1195         struct xfrm_state *x;
1196         u32 mark = m->v & m->m;
1197
1198         hlist_for_each_entry(x, net->xfrm.state_bydst+h, bydst) {
1199                 if (x->props.reqid  != reqid ||
1200                     x->props.mode   != mode ||
1201                     x->props.family != family ||
1202                     x->km.state     != XFRM_STATE_ACQ ||
1203                     x->id.spi       != 0 ||
1204                     x->id.proto     != proto ||
1205                     (mark & x->mark.m) != x->mark.v ||
1206                     !xfrm_addr_equal(&x->id.daddr, daddr, family) ||
1207                     !xfrm_addr_equal(&x->props.saddr, saddr, family))
1208                         continue;
1209
1210                 xfrm_state_hold(x);
1211                 return x;
1212         }
1213
1214         if (!create)
1215                 return NULL;
1216
1217         x = xfrm_state_alloc(net);
1218         if (likely(x)) {
1219                 switch (family) {
1220                 case AF_INET:
1221                         x->sel.daddr.a4 = daddr->a4;
1222                         x->sel.saddr.a4 = saddr->a4;
1223                         x->sel.prefixlen_d = 32;
1224                         x->sel.prefixlen_s = 32;
1225                         x->props.saddr.a4 = saddr->a4;
1226                         x->id.daddr.a4 = daddr->a4;
1227                         break;
1228
1229                 case AF_INET6:
1230                         x->sel.daddr.in6 = daddr->in6;
1231                         x->sel.saddr.in6 = saddr->in6;
1232                         x->sel.prefixlen_d = 128;
1233                         x->sel.prefixlen_s = 128;
1234                         x->props.saddr.in6 = saddr->in6;
1235                         x->id.daddr.in6 = daddr->in6;
1236                         break;
1237                 }
1238
1239                 x->km.state = XFRM_STATE_ACQ;
1240                 x->id.proto = proto;
1241                 x->props.family = family;
1242                 x->props.mode = mode;
1243                 x->props.reqid = reqid;
1244                 x->mark.v = m->v;
1245                 x->mark.m = m->m;
1246                 x->lft.hard_add_expires_seconds = net->xfrm.sysctl_acq_expires;
1247                 xfrm_state_hold(x);
1248                 tasklet_hrtimer_start(&x->mtimer, ktime_set(net->xfrm.sysctl_acq_expires, 0), HRTIMER_MODE_REL);
1249                 list_add(&x->km.all, &net->xfrm.state_all);
1250                 hlist_add_head_rcu(&x->bydst, net->xfrm.state_bydst + h);
1251                 h = xfrm_src_hash(net, daddr, saddr, family);
1252                 hlist_add_head_rcu(&x->bysrc, net->xfrm.state_bysrc + h);
1253
1254                 net->xfrm.state_num++;
1255
1256                 xfrm_hash_grow_check(net, x->bydst.next != NULL);
1257         }
1258
1259         return x;
1260 }
1261
1262 static struct xfrm_state *__xfrm_find_acq_byseq(struct net *net, u32 mark, u32 seq);
1263
1264 int xfrm_state_add(struct xfrm_state *x)
1265 {
1266         struct net *net = xs_net(x);
1267         struct xfrm_state *x1, *to_put;
1268         int family;
1269         int err;
1270         u32 mark = x->mark.v & x->mark.m;
1271         int use_spi = xfrm_id_proto_match(x->id.proto, IPSEC_PROTO_ANY);
1272
1273         family = x->props.family;
1274
1275         to_put = NULL;
1276
1277         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1278
1279         x1 = __xfrm_state_locate(x, use_spi, family);
1280         if (x1) {
1281                 to_put = x1;
1282                 x1 = NULL;
1283                 err = -EEXIST;
1284                 goto out;
1285         }
1286
1287         if (use_spi && x->km.seq) {
1288                 x1 = __xfrm_find_acq_byseq(net, mark, x->km.seq);
1289                 if (x1 && ((x1->id.proto != x->id.proto) ||
1290                     !xfrm_addr_equal(&x1->id.daddr, &x->id.daddr, family))) {
1291                         to_put = x1;
1292                         x1 = NULL;
1293                 }
1294         }
1295
1296         if (use_spi && !x1)
1297                 x1 = __find_acq_core(net, &x->mark, family, x->props.mode,
1298                                      x->props.reqid, x->id.proto,
1299                                      &x->id.daddr, &x->props.saddr, 0);
1300
1301         __xfrm_state_bump_genids(x);
1302         __xfrm_state_insert(x);
1303         err = 0;
1304
1305 out:
1306         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1307
1308         if (x1) {
1309                 xfrm_state_delete(x1);
1310                 xfrm_state_put(x1);
1311         }
1312
1313         if (to_put)
1314                 xfrm_state_put(to_put);
1315
1316         return err;
1317 }
1318 EXPORT_SYMBOL(xfrm_state_add);
1319
1320 #ifdef CONFIG_XFRM_MIGRATE
1321 static struct xfrm_state *xfrm_state_clone(struct xfrm_state *orig,
1322                                            struct xfrm_encap_tmpl *encap)
1323 {
1324         struct net *net = xs_net(orig);
1325         struct xfrm_state *x = xfrm_state_alloc(net);
1326         if (!x)
1327                 goto out;
1328
1329         memcpy(&x->id, &orig->id, sizeof(x->id));
1330         memcpy(&x->sel, &orig->sel, sizeof(x->sel));
1331         memcpy(&x->lft, &orig->lft, sizeof(x->lft));
1332         x->props.mode = orig->props.mode;
1333         x->props.replay_window = orig->props.replay_window;
1334         x->props.reqid = orig->props.reqid;
1335         x->props.family = orig->props.family;
1336         x->props.saddr = orig->props.saddr;
1337
1338         if (orig->aalg) {
1339                 x->aalg = xfrm_algo_auth_clone(orig->aalg);
1340                 if (!x->aalg)
1341                         goto error;
1342         }
1343         x->props.aalgo = orig->props.aalgo;
1344
1345         if (orig->aead) {
1346                 x->aead = xfrm_algo_aead_clone(orig->aead);
1347                 if (!x->aead)
1348                         goto error;
1349         }
1350         if (orig->ealg) {
1351                 x->ealg = xfrm_algo_clone(orig->ealg);
1352                 if (!x->ealg)
1353                         goto error;
1354         }
1355         x->props.ealgo = orig->props.ealgo;
1356
1357         if (orig->calg) {
1358                 x->calg = xfrm_algo_clone(orig->calg);
1359                 if (!x->calg)
1360                         goto error;
1361         }
1362         x->props.calgo = orig->props.calgo;
1363
1364         if (encap || orig->encap) {
1365                 if (encap)
1366                         x->encap = kmemdup(encap, sizeof(*x->encap),
1367                                         GFP_KERNEL);
1368                 else
1369                         x->encap = kmemdup(orig->encap, sizeof(*x->encap),
1370                                         GFP_KERNEL);
1371
1372                 if (!x->encap)
1373                         goto error;
1374         }
1375
1376         if (orig->coaddr) {
1377                 x->coaddr = kmemdup(orig->coaddr, sizeof(*x->coaddr),
1378                                     GFP_KERNEL);
1379                 if (!x->coaddr)
1380                         goto error;
1381         }
1382
1383         if (orig->replay_esn) {
1384                 if (xfrm_replay_clone(x, orig))
1385                         goto error;
1386         }
1387
1388         memcpy(&x->mark, &orig->mark, sizeof(x->mark));
1389
1390         if (xfrm_init_state(x) < 0)
1391                 goto error;
1392
1393         x->props.flags = orig->props.flags;
1394         x->props.extra_flags = orig->props.extra_flags;
1395
1396         x->tfcpad = orig->tfcpad;
1397         x->replay_maxdiff = orig->replay_maxdiff;
1398         x->replay_maxage = orig->replay_maxage;
1399         x->curlft.add_time = orig->curlft.add_time;
1400         x->km.state = orig->km.state;
1401         x->km.seq = orig->km.seq;
1402         x->replay = orig->replay;
1403         x->preplay = orig->preplay;
1404
1405         return x;
1406
1407  error:
1408         xfrm_state_put(x);
1409 out:
1410         return NULL;
1411 }
1412
1413 struct xfrm_state *xfrm_migrate_state_find(struct xfrm_migrate *m, struct net *net)
1414 {
1415         unsigned int h;
1416         struct xfrm_state *x = NULL;
1417
1418         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1419
1420         if (m->reqid) {
1421                 h = xfrm_dst_hash(net, &m->old_daddr, &m->old_saddr,
1422                                   m->reqid, m->old_family);
1423                 hlist_for_each_entry(x, net->xfrm.state_bydst+h, bydst) {
1424                         if (x->props.mode != m->mode ||
1425                             x->id.proto != m->proto)
1426                                 continue;
1427                         if (m->reqid && x->props.reqid != m->reqid)
1428                                 continue;
1429                         if (!xfrm_addr_equal(&x->id.daddr, &m->old_daddr,
1430                                              m->old_family) ||
1431                             !xfrm_addr_equal(&x->props.saddr, &m->old_saddr,
1432                                              m->old_family))
1433                                 continue;
1434                         xfrm_state_hold(x);
1435                         break;
1436                 }
1437         } else {
1438                 h = xfrm_src_hash(net, &m->old_daddr, &m->old_saddr,
1439                                   m->old_family);
1440                 hlist_for_each_entry(x, net->xfrm.state_bysrc+h, bysrc) {
1441                         if (x->props.mode != m->mode ||
1442                             x->id.proto != m->proto)
1443                                 continue;
1444                         if (!xfrm_addr_equal(&x->id.daddr, &m->old_daddr,
1445                                              m->old_family) ||
1446                             !xfrm_addr_equal(&x->props.saddr, &m->old_saddr,
1447                                              m->old_family))
1448                                 continue;
1449                         xfrm_state_hold(x);
1450                         break;
1451                 }
1452         }
1453
1454         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1455
1456         return x;
1457 }
1458 EXPORT_SYMBOL(xfrm_migrate_state_find);
1459
1460 struct xfrm_state *xfrm_state_migrate(struct xfrm_state *x,
1461                                       struct xfrm_migrate *m,
1462                                       struct xfrm_encap_tmpl *encap)
1463 {
1464         struct xfrm_state *xc;
1465
1466         xc = xfrm_state_clone(x, encap);
1467         if (!xc)
1468                 return NULL;
1469
1470         memcpy(&xc->id.daddr, &m->new_daddr, sizeof(xc->id.daddr));
1471         memcpy(&xc->props.saddr, &m->new_saddr, sizeof(xc->props.saddr));
1472
1473         /* add state */
1474         if (xfrm_addr_equal(&x->id.daddr, &m->new_daddr, m->new_family)) {
1475                 /* a care is needed when the destination address of the
1476                    state is to be updated as it is a part of triplet */
1477                 xfrm_state_insert(xc);
1478         } else {
1479                 if (xfrm_state_add(xc) < 0)
1480                         goto error;
1481         }
1482
1483         return xc;
1484 error:
1485         xfrm_state_put(xc);
1486         return NULL;
1487 }
1488 EXPORT_SYMBOL(xfrm_state_migrate);
1489 #endif
1490
1491 int xfrm_state_update(struct xfrm_state *x)
1492 {
1493         struct xfrm_state *x1, *to_put;
1494         int err;
1495         int use_spi = xfrm_id_proto_match(x->id.proto, IPSEC_PROTO_ANY);
1496         struct net *net = xs_net(x);
1497
1498         to_put = NULL;
1499
1500         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1501         x1 = __xfrm_state_locate(x, use_spi, x->props.family);
1502
1503         err = -ESRCH;
1504         if (!x1)
1505                 goto out;
1506
1507         if (xfrm_state_kern(x1)) {
1508                 to_put = x1;
1509                 err = -EEXIST;
1510                 goto out;
1511         }
1512
1513         if (x1->km.state == XFRM_STATE_ACQ) {
1514                 __xfrm_state_insert(x);
1515                 x = NULL;
1516         }
1517         err = 0;
1518
1519 out:
1520         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1521
1522         if (to_put)
1523                 xfrm_state_put(to_put);
1524
1525         if (err)
1526                 return err;
1527
1528         if (!x) {
1529                 xfrm_state_delete(x1);
1530                 xfrm_state_put(x1);
1531                 return 0;
1532         }
1533
1534         err = -EINVAL;
1535         spin_lock_bh(&x1->lock);
1536         if (likely(x1->km.state == XFRM_STATE_VALID)) {
1537                 if (x->encap && x1->encap)
1538                         memcpy(x1->encap, x->encap, sizeof(*x1->encap));
1539                 if (x->coaddr && x1->coaddr) {
1540                         memcpy(x1->coaddr, x->coaddr, sizeof(*x1->coaddr));
1541                 }
1542                 if (!use_spi && memcmp(&x1->sel, &x->sel, sizeof(x1->sel)))
1543                         memcpy(&x1->sel, &x->sel, sizeof(x1->sel));
1544                 memcpy(&x1->lft, &x->lft, sizeof(x1->lft));
1545                 x1->km.dying = 0;
1546
1547                 tasklet_hrtimer_start(&x1->mtimer, ktime_set(1, 0), HRTIMER_MODE_REL);
1548                 if (x1->curlft.use_time)
1549                         xfrm_state_check_expire(x1);
1550
1551                 err = 0;
1552                 x->km.state = XFRM_STATE_DEAD;
1553                 __xfrm_state_put(x);
1554         }
1555         spin_unlock_bh(&x1->lock);
1556
1557         xfrm_state_put(x1);
1558
1559         return err;
1560 }
1561 EXPORT_SYMBOL(xfrm_state_update);
1562
1563 int xfrm_state_check_expire(struct xfrm_state *x)
1564 {
1565         if (!x->curlft.use_time)
1566                 x->curlft.use_time = get_seconds();
1567
1568         if (x->curlft.bytes >= x->lft.hard_byte_limit ||
1569             x->curlft.packets >= x->lft.hard_packet_limit) {
1570                 x->km.state = XFRM_STATE_EXPIRED;
1571                 tasklet_hrtimer_start(&x->mtimer, 0, HRTIMER_MODE_REL);
1572                 return -EINVAL;
1573         }
1574
1575         if (!x->km.dying &&
1576             (x->curlft.bytes >= x->lft.soft_byte_limit ||
1577              x->curlft.packets >= x->lft.soft_packet_limit)) {
1578                 x->km.dying = 1;
1579                 km_state_expired(x, 0, 0);
1580         }
1581         return 0;
1582 }
1583 EXPORT_SYMBOL(xfrm_state_check_expire);
1584
1585 struct xfrm_state *
1586 xfrm_state_lookup(struct net *net, u32 mark, const xfrm_address_t *daddr, __be32 spi,
1587                   u8 proto, unsigned short family)
1588 {
1589         struct xfrm_state *x;
1590
1591         rcu_read_lock();
1592         x = __xfrm_state_lookup(net, mark, daddr, spi, proto, family);
1593         rcu_read_unlock();
1594         return x;
1595 }
1596 EXPORT_SYMBOL(xfrm_state_lookup);
1597
1598 struct xfrm_state *
1599 xfrm_state_lookup_byaddr(struct net *net, u32 mark,
1600                          const xfrm_address_t *daddr, const xfrm_address_t *saddr,
1601                          u8 proto, unsigned short family)
1602 {
1603         struct xfrm_state *x;
1604
1605         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1606         x = __xfrm_state_lookup_byaddr(net, mark, daddr, saddr, proto, family);
1607         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1608         return x;
1609 }
1610 EXPORT_SYMBOL(xfrm_state_lookup_byaddr);
1611
1612 struct xfrm_state *
1613 xfrm_find_acq(struct net *net, const struct xfrm_mark *mark, u8 mode, u32 reqid,
1614               u8 proto, const xfrm_address_t *daddr,
1615               const xfrm_address_t *saddr, int create, unsigned short family)
1616 {
1617         struct xfrm_state *x;
1618
1619         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1620         x = __find_acq_core(net, mark, family, mode, reqid, proto, daddr, saddr, create);
1621         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1622
1623         return x;
1624 }
1625 EXPORT_SYMBOL(xfrm_find_acq);
1626
1627 #ifdef CONFIG_XFRM_SUB_POLICY
1628 int
1629 xfrm_tmpl_sort(struct xfrm_tmpl **dst, struct xfrm_tmpl **src, int n,
1630                unsigned short family, struct net *net)
1631 {
1632         int i;
1633         int err = 0;
1634         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
1635         if (!afinfo)
1636                 return -EAFNOSUPPORT;
1637
1638         spin_lock_bh(&net->xfrm.xfrm_state_lock); /*FIXME*/
1639         if (afinfo->tmpl_sort)
1640                 err = afinfo->tmpl_sort(dst, src, n);
1641         else
1642                 for (i = 0; i < n; i++)
1643                         dst[i] = src[i];
1644         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1645         rcu_read_unlock();
1646         return err;
1647 }
1648 EXPORT_SYMBOL(xfrm_tmpl_sort);
1649
1650 int
1651 xfrm_state_sort(struct xfrm_state **dst, struct xfrm_state **src, int n,
1652                 unsigned short family)
1653 {
1654         int i;
1655         int err = 0;
1656         struct xfrm_state_afinfo *afinfo = xfrm_state_get_afinfo(family);
1657         struct net *net = xs_net(*src);
1658
1659         if (!afinfo)
1660                 return -EAFNOSUPPORT;
1661
1662         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1663         if (afinfo->state_sort)
1664                 err = afinfo->state_sort(dst, src, n);
1665         else
1666                 for (i = 0; i < n; i++)
1667                         dst[i] = src[i];
1668         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1669         rcu_read_unlock();
1670         return err;
1671 }
1672 EXPORT_SYMBOL(xfrm_state_sort);
1673 #endif
1674
1675 /* Silly enough, but I'm lazy to build resolution list */
1676
1677 static struct xfrm_state *__xfrm_find_acq_byseq(struct net *net, u32 mark, u32 seq)
1678 {
1679         int i;
1680
1681         for (i = 0; i <= net->xfrm.state_hmask; i++) {
1682                 struct xfrm_state *x;
1683
1684                 hlist_for_each_entry(x, net->xfrm.state_bydst+i, bydst) {
1685                         if (x->km.seq == seq &&
1686                             (mark & x->mark.m) == x->mark.v &&
1687                             x->km.state == XFRM_STATE_ACQ) {
1688                                 xfrm_state_hold(x);
1689                                 return x;
1690                         }
1691                 }
1692         }
1693         return NULL;
1694 }
1695
1696 struct xfrm_state *xfrm_find_acq_byseq(struct net *net, u32 mark, u32 seq)
1697 {
1698         struct xfrm_state *x;
1699
1700         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1701         x = __xfrm_find_acq_byseq(net, mark, seq);
1702         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1703         return x;
1704 }
1705 EXPORT_SYMBOL(xfrm_find_acq_byseq);
1706
1707 u32 xfrm_get_acqseq(void)
1708 {
1709         u32 res;
1710         static atomic_t acqseq;
1711
1712         do {
1713                 res = atomic_inc_return(&acqseq);
1714         } while (!res);
1715
1716         return res;
1717 }
1718 EXPORT_SYMBOL(xfrm_get_acqseq);
1719
1720 int verify_spi_info(u8 proto, u32 min, u32 max)
1721 {
1722         switch (proto) {
1723         case IPPROTO_AH:
1724         case IPPROTO_ESP:
1725                 break;
1726
1727         case IPPROTO_COMP:
1728                 /* IPCOMP spi is 16-bits. */
1729                 if (max >= 0x10000)
1730                         return -EINVAL;
1731                 break;
1732
1733         default:
1734                 return -EINVAL;
1735         }
1736
1737         if (min > max)
1738                 return -EINVAL;
1739
1740         return 0;
1741 }
1742 EXPORT_SYMBOL(verify_spi_info);
1743
1744 int xfrm_alloc_spi(struct xfrm_state *x, u32 low, u32 high)
1745 {
1746         struct net *net = xs_net(x);
1747         unsigned int h;
1748         struct xfrm_state *x0;
1749         int err = -ENOENT;
1750         __be32 minspi = htonl(low);
1751         __be32 maxspi = htonl(high);
1752         u32 mark = x->mark.v & x->mark.m;
1753
1754         spin_lock_bh(&x->lock);
1755         if (x->km.state == XFRM_STATE_DEAD)
1756                 goto unlock;
1757
1758         err = 0;
1759         if (x->id.spi)
1760                 goto unlock;
1761
1762         err = -ENOENT;
1763
1764         if (minspi == maxspi) {
1765                 x0 = xfrm_state_lookup(net, mark, &x->id.daddr, minspi, x->id.proto, x->props.family);
1766                 if (x0) {
1767                         xfrm_state_put(x0);
1768                         goto unlock;
1769                 }
1770                 x->id.spi = minspi;
1771         } else {
1772                 u32 spi = 0;
1773                 for (h = 0; h < high-low+1; h++) {
1774                         spi = low + prandom_u32()%(high-low+1);
1775                         x0 = xfrm_state_lookup(net, mark, &x->id.daddr, htonl(spi), x->id.proto, x->props.family);
1776                         if (x0 == NULL) {
1777                                 x->id.spi = htonl(spi);
1778                                 break;
1779                         }
1780                         xfrm_state_put(x0);
1781                 }
1782         }
1783         if (x->id.spi) {
1784                 spin_lock_bh(&net->xfrm.xfrm_state_lock);
1785                 h = xfrm_spi_hash(net, &x->id.daddr, x->id.spi, x->id.proto, x->props.family);
1786                 hlist_add_head_rcu(&x->byspi, net->xfrm.state_byspi + h);
1787                 spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1788
1789                 err = 0;
1790         }
1791
1792 unlock:
1793         spin_unlock_bh(&x->lock);
1794
1795         return err;
1796 }
1797 EXPORT_SYMBOL(xfrm_alloc_spi);
1798
1799 static bool __xfrm_state_filter_match(struct xfrm_state *x,
1800                                       struct xfrm_address_filter *filter)
1801 {
1802         if (filter) {
1803                 if ((filter->family == AF_INET ||
1804                      filter->family == AF_INET6) &&
1805                     x->props.family != filter->family)
1806                         return false;
1807
1808                 return addr_match(&x->props.saddr, &filter->saddr,
1809                                   filter->splen) &&
1810                        addr_match(&x->id.daddr, &filter->daddr,
1811                                   filter->dplen);
1812         }
1813         return true;
1814 }
1815
1816 int xfrm_state_walk(struct net *net, struct xfrm_state_walk *walk,
1817                     int (*func)(struct xfrm_state *, int, void*),
1818                     void *data)
1819 {
1820         struct xfrm_state *state;
1821         struct xfrm_state_walk *x;
1822         int err = 0;
1823
1824         if (walk->seq != 0 && list_empty(&walk->all))
1825                 return 0;
1826
1827         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1828         if (list_empty(&walk->all))
1829                 x = list_first_entry(&net->xfrm.state_all, struct xfrm_state_walk, all);
1830         else
1831                 x = list_first_entry(&walk->all, struct xfrm_state_walk, all);
1832         list_for_each_entry_from(x, &net->xfrm.state_all, all) {
1833                 if (x->state == XFRM_STATE_DEAD)
1834                         continue;
1835                 state = container_of(x, struct xfrm_state, km);
1836                 if (!xfrm_id_proto_match(state->id.proto, walk->proto))
1837                         continue;
1838                 if (!__xfrm_state_filter_match(state, walk->filter))
1839                         continue;
1840                 err = func(state, walk->seq, data);
1841                 if (err) {
1842                         list_move_tail(&walk->all, &x->all);
1843                         goto out;
1844                 }
1845                 walk->seq++;
1846         }
1847         if (walk->seq == 0) {
1848                 err = -ENOENT;
1849                 goto out;
1850         }
1851         list_del_init(&walk->all);
1852 out:
1853         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1854         return err;
1855 }
1856 EXPORT_SYMBOL(xfrm_state_walk);
1857
1858 void xfrm_state_walk_init(struct xfrm_state_walk *walk, u8 proto,
1859                           struct xfrm_address_filter *filter)
1860 {
1861         INIT_LIST_HEAD(&walk->all);
1862         walk->proto = proto;
1863         walk->state = XFRM_STATE_DEAD;
1864         walk->seq = 0;
1865         walk->filter = filter;
1866 }
1867 EXPORT_SYMBOL(xfrm_state_walk_init);
1868
1869 void xfrm_state_walk_done(struct xfrm_state_walk *walk, struct net *net)
1870 {
1871         kfree(walk->filter);
1872
1873         if (list_empty(&walk->all))
1874                 return;
1875
1876         spin_lock_bh(&net->xfrm.xfrm_state_lock);
1877         list_del(&walk->all);
1878         spin_unlock_bh(&net->xfrm.xfrm_state_lock);
1879 }
1880 EXPORT_SYMBOL(xfrm_state_walk_done);
1881
1882 static void xfrm_replay_timer_handler(unsigned long data)
1883 {
1884         struct xfrm_state *x = (struct xfrm_state *)data;
1885
1886         spin_lock(&x->lock);
1887
1888         if (x->km.state == XFRM_STATE_VALID) {
1889                 if (xfrm_aevent_is_on(xs_net(x)))
1890                         x->repl->notify(x, XFRM_REPLAY_TIMEOUT);
1891                 else
1892                         x->xflags |= XFRM_TIME_DEFER;
1893         }
1894
1895         spin_unlock(&x->lock);
1896 }
1897
1898 static LIST_HEAD(xfrm_km_list);
1899
1900 void km_policy_notify(struct xfrm_policy *xp, int dir, const struct km_event *c)
1901 {
1902         struct xfrm_mgr *km;
1903
1904         rcu_read_lock();
1905         list_for_each_entry_rcu(km, &xfrm_km_list, list)
1906                 if (km->notify_policy)
1907                         km->notify_policy(xp, dir, c);
1908         rcu_read_unlock();
1909 }
1910
1911 void km_state_notify(struct xfrm_state *x, const struct km_event *c)
1912 {
1913         struct xfrm_mgr *km;
1914         rcu_read_lock();
1915         list_for_each_entry_rcu(km, &xfrm_km_list, list)
1916                 if (km->notify)
1917                         km->notify(x, c);
1918         rcu_read_unlock();
1919 }
1920
1921 EXPORT_SYMBOL(km_policy_notify);
1922 EXPORT_SYMBOL(km_state_notify);
1923
1924 void km_state_expired(struct xfrm_state *x, int hard, u32 portid)
1925 {
1926         struct km_event c;
1927
1928         c.data.hard = hard;
1929         c.portid = portid;
1930         c.event = XFRM_MSG_EXPIRE;
1931         km_state_notify(x, &c);
1932 }
1933
1934 EXPORT_SYMBOL(km_state_expired);
1935 /*
1936  * We send to all registered managers regardless of failure
1937  * We are happy with one success
1938 */
1939 int km_query(struct xfrm_state *x, struct xfrm_tmpl *t, struct xfrm_policy *pol)
1940 {
1941         int err = -EINVAL, acqret;
1942         struct xfrm_mgr *km;
1943
1944         rcu_read_lock();
1945         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
1946                 acqret = km->acquire(x, t, pol);
1947                 if (!acqret)
1948                         err = acqret;
1949         }
1950         rcu_read_unlock();
1951         return err;
1952 }
1953 EXPORT_SYMBOL(km_query);
1954
1955 int km_new_mapping(struct xfrm_state *x, xfrm_address_t *ipaddr, __be16 sport)
1956 {
1957         int err = -EINVAL;
1958         struct xfrm_mgr *km;
1959
1960         rcu_read_lock();
1961         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
1962                 if (km->new_mapping)
1963                         err = km->new_mapping(x, ipaddr, sport);
1964                 if (!err)
1965                         break;
1966         }
1967         rcu_read_unlock();
1968         return err;
1969 }
1970 EXPORT_SYMBOL(km_new_mapping);
1971
1972 void km_policy_expired(struct xfrm_policy *pol, int dir, int hard, u32 portid)
1973 {
1974         struct km_event c;
1975
1976         c.data.hard = hard;
1977         c.portid = portid;
1978         c.event = XFRM_MSG_POLEXPIRE;
1979         km_policy_notify(pol, dir, &c);
1980 }
1981 EXPORT_SYMBOL(km_policy_expired);
1982
1983 #ifdef CONFIG_XFRM_MIGRATE
1984 int km_migrate(const struct xfrm_selector *sel, u8 dir, u8 type,
1985                const struct xfrm_migrate *m, int num_migrate,
1986                const struct xfrm_kmaddress *k,
1987                const struct xfrm_encap_tmpl *encap)
1988 {
1989         int err = -EINVAL;
1990         int ret;
1991         struct xfrm_mgr *km;
1992
1993         rcu_read_lock();
1994         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
1995                 if (km->migrate) {
1996                         ret = km->migrate(sel, dir, type, m, num_migrate, k,
1997                                           encap);
1998                         if (!ret)
1999                                 err = ret;
2000                 }
2001         }
2002         rcu_read_unlock();
2003         return err;
2004 }
2005 EXPORT_SYMBOL(km_migrate);
2006 #endif
2007
2008 int km_report(struct net *net, u8 proto, struct xfrm_selector *sel, xfrm_address_t *addr)
2009 {
2010         int err = -EINVAL;
2011         int ret;
2012         struct xfrm_mgr *km;
2013
2014         rcu_read_lock();
2015         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
2016                 if (km->report) {
2017                         ret = km->report(net, proto, sel, addr);
2018                         if (!ret)
2019                                 err = ret;
2020                 }
2021         }
2022         rcu_read_unlock();
2023         return err;
2024 }
2025 EXPORT_SYMBOL(km_report);
2026
2027 bool km_is_alive(const struct km_event *c)
2028 {
2029         struct xfrm_mgr *km;
2030         bool is_alive = false;
2031
2032         rcu_read_lock();
2033         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
2034                 if (km->is_alive && km->is_alive(c)) {
2035                         is_alive = true;
2036                         break;
2037                 }
2038         }
2039         rcu_read_unlock();
2040
2041         return is_alive;
2042 }
2043 EXPORT_SYMBOL(km_is_alive);
2044
2045 int xfrm_user_policy(struct sock *sk, int optname, u8 __user *optval, int optlen)
2046 {
2047         int err;
2048         u8 *data;
2049         struct xfrm_mgr *km;
2050         struct xfrm_policy *pol = NULL;
2051
2052         if (optlen <= 0 || optlen > PAGE_SIZE)
2053                 return -EMSGSIZE;
2054
2055         data = memdup_user(optval, optlen);
2056         if (IS_ERR(data))
2057                 return PTR_ERR(data);
2058
2059         err = -EINVAL;
2060         rcu_read_lock();
2061         list_for_each_entry_rcu(km, &xfrm_km_list, list) {
2062                 pol = km->compile_policy(sk, optname, data,
2063                                          optlen, &err);
2064                 if (err >= 0)
2065                         break;
2066         }
2067         rcu_read_unlock();
2068
2069         if (err >= 0) {
2070                 xfrm_sk_policy_insert(sk, err, pol);
2071                 xfrm_pol_put(pol);
2072                 err = 0;
2073         }
2074
2075         kfree(data);
2076         return err;
2077 }
2078 EXPORT_SYMBOL(xfrm_user_policy);
2079
2080 static DEFINE_SPINLOCK(xfrm_km_lock);
2081
2082 int xfrm_register_km(struct xfrm_mgr *km)
2083 {
2084         spin_lock_bh(&xfrm_km_lock);
2085         list_add_tail_rcu(&km->list, &xfrm_km_list);
2086         spin_unlock_bh(&xfrm_km_lock);
2087         return 0;
2088 }
2089 EXPORT_SYMBOL(xfrm_register_km);
2090
2091 int xfrm_unregister_km(struct xfrm_mgr *km)
2092 {
2093         spin_lock_bh(&xfrm_km_lock);
2094         list_del_rcu(&km->list);
2095         spin_unlock_bh(&xfrm_km_lock);
2096         synchronize_rcu();
2097         return 0;
2098 }
2099 EXPORT_SYMBOL(xfrm_unregister_km);
2100
2101 int xfrm_state_register_afinfo(struct xfrm_state_afinfo *afinfo)
2102 {
2103         int err = 0;
2104
2105         if (WARN_ON(afinfo->family >= NPROTO))
2106                 return -EAFNOSUPPORT;
2107
2108         spin_lock_bh(&xfrm_state_afinfo_lock);
2109         if (unlikely(xfrm_state_afinfo[afinfo->family] != NULL))
2110                 err = -EEXIST;
2111         else
2112                 rcu_assign_pointer(xfrm_state_afinfo[afinfo->family], afinfo);
2113         spin_unlock_bh(&xfrm_state_afinfo_lock);
2114         return err;
2115 }
2116 EXPORT_SYMBOL(xfrm_state_register_afinfo);
2117
2118 int xfrm_state_unregister_afinfo(struct xfrm_state_afinfo *afinfo)
2119 {
2120         int err = 0, family = afinfo->family;
2121
2122         if (WARN_ON(family >= NPROTO))
2123                 return -EAFNOSUPPORT;
2124
2125         spin_lock_bh(&xfrm_state_afinfo_lock);
2126         if (likely(xfrm_state_afinfo[afinfo->family] != NULL)) {
2127                 if (rcu_access_pointer(xfrm_state_afinfo[family]) != afinfo)
2128                         err = -EINVAL;
2129                 else
2130                         RCU_INIT_POINTER(xfrm_state_afinfo[afinfo->family], NULL);
2131         }
2132         spin_unlock_bh(&xfrm_state_afinfo_lock);
2133         synchronize_rcu();
2134         return err;
2135 }
2136 EXPORT_SYMBOL(xfrm_state_unregister_afinfo);
2137
2138 struct xfrm_state_afinfo *xfrm_state_afinfo_get_rcu(unsigned int family)
2139 {
2140         if (unlikely(family >= NPROTO))
2141                 return NULL;
2142
2143         return rcu_dereference(xfrm_state_afinfo[family]);
2144 }
2145
2146 struct xfrm_state_afinfo *xfrm_state_get_afinfo(unsigned int family)
2147 {
2148         struct xfrm_state_afinfo *afinfo;
2149         if (unlikely(family >= NPROTO))
2150                 return NULL;
2151         rcu_read_lock();
2152         afinfo = rcu_dereference(xfrm_state_afinfo[family]);
2153         if (unlikely(!afinfo))
2154                 rcu_read_unlock();
2155         return afinfo;
2156 }
2157
2158 /* Temporarily located here until net/xfrm/xfrm_tunnel.c is created */
2159 void xfrm_state_delete_tunnel(struct xfrm_state *x)
2160 {
2161         if (x->tunnel) {
2162                 struct xfrm_state *t = x->tunnel;
2163
2164                 if (atomic_read(&t->tunnel_users) == 2)
2165                         xfrm_state_delete(t);
2166                 atomic_dec(&t->tunnel_users);
2167                 xfrm_state_put(t);
2168                 x->tunnel = NULL;
2169         }
2170 }
2171 EXPORT_SYMBOL(xfrm_state_delete_tunnel);
2172
2173 int xfrm_state_mtu(struct xfrm_state *x, int mtu)
2174 {
2175         const struct xfrm_type *type = READ_ONCE(x->type);
2176
2177         if (x->km.state == XFRM_STATE_VALID &&
2178             type && type->get_mtu)
2179                 return type->get_mtu(x, mtu);
2180
2181         return mtu - x->props.header_len;
2182 }
2183
2184 int __xfrm_init_state(struct xfrm_state *x, bool init_replay, bool offload)
2185 {
2186         struct xfrm_state_afinfo *afinfo;
2187         struct xfrm_mode *inner_mode;
2188         int family = x->props.family;
2189         int err;
2190
2191         err = -EAFNOSUPPORT;
2192         afinfo = xfrm_state_get_afinfo(family);
2193         if (!afinfo)
2194                 goto error;
2195
2196         err = 0;
2197         if (afinfo->init_flags)
2198                 err = afinfo->init_flags(x);
2199
2200         rcu_read_unlock();
2201
2202         if (err)
2203                 goto error;
2204
2205         err = -EPROTONOSUPPORT;
2206
2207         if (x->sel.family != AF_UNSPEC) {
2208                 inner_mode = xfrm_get_mode(x->props.mode, x->sel.family);
2209                 if (inner_mode == NULL)
2210                         goto error;
2211
2212                 if (!(inner_mode->flags & XFRM_MODE_FLAG_TUNNEL) &&
2213                     family != x->sel.family) {
2214                         xfrm_put_mode(inner_mode);
2215                         goto error;
2216                 }
2217
2218                 x->inner_mode = inner_mode;
2219         } else {
2220                 struct xfrm_mode *inner_mode_iaf;
2221                 int iafamily = AF_INET;
2222
2223                 inner_mode = xfrm_get_mode(x->props.mode, x->props.family);
2224                 if (inner_mode == NULL)
2225                         goto error;
2226
2227                 if (!(inner_mode->flags & XFRM_MODE_FLAG_TUNNEL)) {
2228                         xfrm_put_mode(inner_mode);
2229                         goto error;
2230                 }
2231                 x->inner_mode = inner_mode;
2232
2233                 if (x->props.family == AF_INET)
2234                         iafamily = AF_INET6;
2235
2236                 inner_mode_iaf = xfrm_get_mode(x->props.mode, iafamily);
2237                 if (inner_mode_iaf) {
2238                         if (inner_mode_iaf->flags & XFRM_MODE_FLAG_TUNNEL)
2239                                 x->inner_mode_iaf = inner_mode_iaf;
2240                         else
2241                                 xfrm_put_mode(inner_mode_iaf);
2242                 }
2243         }
2244
2245         x->type = xfrm_get_type(x->id.proto, family);
2246         if (x->type == NULL)
2247                 goto error;
2248
2249         x->type_offload = xfrm_get_type_offload(x->id.proto, family, offload);
2250
2251         err = x->type->init_state(x);
2252         if (err)
2253                 goto error;
2254
2255         x->outer_mode = xfrm_get_mode(x->props.mode, family);
2256         if (x->outer_mode == NULL) {
2257                 err = -EPROTONOSUPPORT;
2258                 goto error;
2259         }
2260
2261         if (init_replay) {
2262                 err = xfrm_init_replay(x);
2263                 if (err)
2264                         goto error;
2265         }
2266
2267         x->km.state = XFRM_STATE_VALID;
2268
2269 error:
2270         return err;
2271 }
2272
2273 EXPORT_SYMBOL(__xfrm_init_state);
2274
2275 int xfrm_init_state(struct xfrm_state *x)
2276 {
2277         return __xfrm_init_state(x, true, false);
2278 }
2279
2280 EXPORT_SYMBOL(xfrm_init_state);
2281
2282 int __net_init xfrm_state_init(struct net *net)
2283 {
2284         unsigned int sz;
2285
2286         INIT_LIST_HEAD(&net->xfrm.state_all);
2287
2288         sz = sizeof(struct hlist_head) * 8;
2289
2290         net->xfrm.state_bydst = xfrm_hash_alloc(sz);
2291         if (!net->xfrm.state_bydst)
2292                 goto out_bydst;
2293         net->xfrm.state_bysrc = xfrm_hash_alloc(sz);
2294         if (!net->xfrm.state_bysrc)
2295                 goto out_bysrc;
2296         net->xfrm.state_byspi = xfrm_hash_alloc(sz);
2297         if (!net->xfrm.state_byspi)
2298                 goto out_byspi;
2299         net->xfrm.state_hmask = ((sz / sizeof(struct hlist_head)) - 1);
2300
2301         net->xfrm.state_num = 0;
2302         INIT_WORK(&net->xfrm.state_hash_work, xfrm_hash_resize);
2303         spin_lock_init(&net->xfrm.xfrm_state_lock);
2304         return 0;
2305
2306 out_byspi:
2307         xfrm_hash_free(net->xfrm.state_bysrc, sz);
2308 out_bysrc:
2309         xfrm_hash_free(net->xfrm.state_bydst, sz);
2310 out_bydst:
2311         return -ENOMEM;
2312 }
2313
2314 void xfrm_state_fini(struct net *net)
2315 {
2316         unsigned int sz;
2317
2318         flush_work(&net->xfrm.state_hash_work);
2319         xfrm_state_flush(net, IPSEC_PROTO_ANY, false);
2320         flush_work(&xfrm_state_gc_work);
2321
2322         WARN_ON(!list_empty(&net->xfrm.state_all));
2323
2324         sz = (net->xfrm.state_hmask + 1) * sizeof(struct hlist_head);
2325         WARN_ON(!hlist_empty(net->xfrm.state_byspi));
2326         xfrm_hash_free(net->xfrm.state_byspi, sz);
2327         WARN_ON(!hlist_empty(net->xfrm.state_bysrc));
2328         xfrm_hash_free(net->xfrm.state_bysrc, sz);
2329         WARN_ON(!hlist_empty(net->xfrm.state_bydst));
2330         xfrm_hash_free(net->xfrm.state_bydst, sz);
2331 }
2332
2333 #ifdef CONFIG_AUDITSYSCALL
2334 static void xfrm_audit_helper_sainfo(struct xfrm_state *x,
2335                                      struct audit_buffer *audit_buf)
2336 {
2337         struct xfrm_sec_ctx *ctx = x->security;
2338         u32 spi = ntohl(x->id.spi);
2339
2340         if (ctx)
2341                 audit_log_format(audit_buf, " sec_alg=%u sec_doi=%u sec_obj=%s",
2342                                  ctx->ctx_alg, ctx->ctx_doi, ctx->ctx_str);
2343
2344         switch (x->props.family) {
2345         case AF_INET:
2346                 audit_log_format(audit_buf, " src=%pI4 dst=%pI4",
2347                                  &x->props.saddr.a4, &x->id.daddr.a4);
2348                 break;
2349         case AF_INET6:
2350                 audit_log_format(audit_buf, " src=%pI6 dst=%pI6",
2351                                  x->props.saddr.a6, x->id.daddr.a6);
2352                 break;
2353         }
2354
2355         audit_log_format(audit_buf, " spi=%u(0x%x)", spi, spi);
2356 }
2357
2358 static void xfrm_audit_helper_pktinfo(struct sk_buff *skb, u16 family,
2359                                       struct audit_buffer *audit_buf)
2360 {
2361         const struct iphdr *iph4;
2362         const struct ipv6hdr *iph6;
2363
2364         switch (family) {
2365         case AF_INET:
2366                 iph4 = ip_hdr(skb);
2367                 audit_log_format(audit_buf, " src=%pI4 dst=%pI4",
2368                                  &iph4->saddr, &iph4->daddr);
2369                 break;
2370         case AF_INET6:
2371                 iph6 = ipv6_hdr(skb);
2372                 audit_log_format(audit_buf,
2373                                  " src=%pI6 dst=%pI6 flowlbl=0x%x%02x%02x",
2374                                  &iph6->saddr, &iph6->daddr,
2375                                  iph6->flow_lbl[0] & 0x0f,
2376                                  iph6->flow_lbl[1],
2377                                  iph6->flow_lbl[2]);
2378                 break;
2379         }
2380 }
2381
2382 void xfrm_audit_state_add(struct xfrm_state *x, int result, bool task_valid)
2383 {
2384         struct audit_buffer *audit_buf;
2385
2386         audit_buf = xfrm_audit_start("SAD-add");
2387         if (audit_buf == NULL)
2388                 return;
2389         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
2390         xfrm_audit_helper_sainfo(x, audit_buf);
2391         audit_log_format(audit_buf, " res=%u", result);
2392         audit_log_end(audit_buf);
2393 }
2394 EXPORT_SYMBOL_GPL(xfrm_audit_state_add);
2395
2396 void xfrm_audit_state_delete(struct xfrm_state *x, int result, bool task_valid)
2397 {
2398         struct audit_buffer *audit_buf;
2399
2400         audit_buf = xfrm_audit_start("SAD-delete");
2401         if (audit_buf == NULL)
2402                 return;
2403         xfrm_audit_helper_usrinfo(task_valid, audit_buf);
2404         xfrm_audit_helper_sainfo(x, audit_buf);
2405         audit_log_format(audit_buf, " res=%u", result);
2406         audit_log_end(audit_buf);
2407 }
2408 EXPORT_SYMBOL_GPL(xfrm_audit_state_delete);
2409
2410 void xfrm_audit_state_replay_overflow(struct xfrm_state *x,
2411                                       struct sk_buff *skb)
2412 {
2413         struct audit_buffer *audit_buf;
2414         u32 spi;
2415
2416         audit_buf = xfrm_audit_start("SA-replay-overflow");
2417         if (audit_buf == NULL)
2418                 return;
2419         xfrm_audit_helper_pktinfo(skb, x->props.family, audit_buf);
2420         /* don't record the sequence number because it's inherent in this kind
2421          * of audit message */
2422         spi = ntohl(x->id.spi);
2423         audit_log_format(audit_buf, " spi=%u(0x%x)", spi, spi);
2424         audit_log_end(audit_buf);
2425 }
2426 EXPORT_SYMBOL_GPL(xfrm_audit_state_replay_overflow);
2427
2428 void xfrm_audit_state_replay(struct xfrm_state *x,
2429                              struct sk_buff *skb, __be32 net_seq)
2430 {
2431         struct audit_buffer *audit_buf;
2432         u32 spi;
2433
2434         audit_buf = xfrm_audit_start("SA-replayed-pkt");
2435         if (audit_buf == NULL)
2436                 return;
2437         xfrm_audit_helper_pktinfo(skb, x->props.family, audit_buf);
2438         spi = ntohl(x->id.spi);
2439         audit_log_format(audit_buf, " spi=%u(0x%x) seqno=%u",
2440                          spi, spi, ntohl(net_seq));
2441         audit_log_end(audit_buf);
2442 }
2443 EXPORT_SYMBOL_GPL(xfrm_audit_state_replay);
2444
2445 void xfrm_audit_state_notfound_simple(struct sk_buff *skb, u16 family)
2446 {
2447         struct audit_buffer *audit_buf;
2448
2449         audit_buf = xfrm_audit_start("SA-notfound");
2450         if (audit_buf == NULL)
2451                 return;
2452         xfrm_audit_helper_pktinfo(skb, family, audit_buf);
2453         audit_log_end(audit_buf);
2454 }
2455 EXPORT_SYMBOL_GPL(xfrm_audit_state_notfound_simple);
2456
2457 void xfrm_audit_state_notfound(struct sk_buff *skb, u16 family,
2458                                __be32 net_spi, __be32 net_seq)
2459 {
2460         struct audit_buffer *audit_buf;
2461         u32 spi;
2462
2463         audit_buf = xfrm_audit_start("SA-notfound");
2464         if (audit_buf == NULL)
2465                 return;
2466         xfrm_audit_helper_pktinfo(skb, family, audit_buf);
2467         spi = ntohl(net_spi);
2468         audit_log_format(audit_buf, " spi=%u(0x%x) seqno=%u",
2469                          spi, spi, ntohl(net_seq));
2470         audit_log_end(audit_buf);
2471 }
2472 EXPORT_SYMBOL_GPL(xfrm_audit_state_notfound);
2473
2474 void xfrm_audit_state_icvfail(struct xfrm_state *x,
2475                               struct sk_buff *skb, u8 proto)
2476 {
2477         struct audit_buffer *audit_buf;
2478         __be32 net_spi;
2479         __be32 net_seq;
2480
2481         audit_buf = xfrm_audit_start("SA-icv-failure");
2482         if (audit_buf == NULL)
2483                 return;
2484         xfrm_audit_helper_pktinfo(skb, x->props.family, audit_buf);
2485         if (xfrm_parse_spi(skb, proto, &net_spi, &net_seq) == 0) {
2486                 u32 spi = ntohl(net_spi);
2487                 audit_log_format(audit_buf, " spi=%u(0x%x) seqno=%u",
2488                                  spi, spi, ntohl(net_seq));
2489         }
2490         audit_log_end(audit_buf);
2491 }
2492 EXPORT_SYMBOL_GPL(xfrm_audit_state_icvfail);
2493 #endif /* CONFIG_AUDITSYSCALL */